JP2003527649A - System and method for database similarity join - Google Patents

Abstract

(57) [Summary] What is disclosed is a system and method for organizing information, in which characteristics regarding an entity are inferred from characteristics of a similar entity. This is referred to herein as "fuzzy-like binding" and is illustrated using drug-like binding. The disclosed systems and methods are particularly useful for compound analysis in the area of drug development.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION

[0002]

FIELD OF THE INVENTION

The present invention relates generally to information databases, and more particularly to database-like joins, and more particularly to systems and methods for organizing information so that overall features can be inferred from similar overall features. This is referred to herein as a "fuzzy" like bond and is exemplified using a drug like bond.

[0003]

[Background information]

[0004]

[Development of drug candidates]

Chemists, biologists and other users always make and test sets of compounds to investigate and substantiate hypotheses. In the process, users often seek to obtain compounds that exhibit certain characteristics or behave according to certain metrics, and seek for synthetic compounds with similar characteristics or behavior patterns. I have something to do.

The method of looking for a certain compound of market value usually starts with extensive selection and testing. An example of this is a global screen typically used in the initial stages of drug discovery. Although drug discovery is used as an example, the same type of process is used in pesticide discovery and material science research, and other related disciplines.

In high throughput screening (HTS), the number of compounds tested and tested for promising biological responses often ranges from 50,000 to 500,000 or more. The ultimate goal is to find some minor set of the large set of compounds that are active in biological screens, and these compounds as “leads” that can be further developed into final drug candidates. Is to treat the compound. The first library of compounds tested represents many different types of drugs.

The drugs in the initial library can be obtained from several sources, including those developed in-house by routine synthesis, commercial availability, chemical combination, and extraction of natural products. These compounds are typically used in microtiter plates (micr).
placed in the o-titer plate). Typical formats for plates include 96 and 384 well plates, but 1536 and 3
Tends towards higher density plates such as 456 well plates. These plates are commonly manipulated by robots to perform biological screens.

The screen itself is usually based on biological receptors. Receptors are sequestered so that binding to the receptor can be measured to some extent more directly, or cell lines are provided to give a detectable response when the receptor is modulated by a potential drug lead. line) is either designed.

[0009] While the vast majority of initial libraries contain thousands of compounds, the more extensive libraries show a myriad of simple subsets of potential drug structures that may have "mist-like" properties. The total number of compounds available on the market is now estimated to be limited to a total of about one million species.

The table of compounds to be screened can be randomly selected from these available ones, or some intuitive “chemist” or “biologist” according to a particular scheme.
Often has a "bias". This bias is often beneficial to programs where chemists often have a unique insight into the types of chemicals that can lead to useful drug candidates. However, due to any bias, sometimes intuitive methods can overlook potential new drugs.

In the last few years, the technological trend has been to make selections based on the dissimilarity of compounds within the final selected set. This process is intended to ensure that many broad classes of compounds are tested. Dissimilarity measure (diversity metric)
And methods of selecting dissimilarities have been much discussed, but these always rely on a measure of "similarity" between two compounds. While it is common to choose compounds that are as different from each other as possible, this often leads to the selection of the most chemically "unique" compounds in the set. Therefore, this method may lead to missing potential active lead compounds.

In guiding these studies, researchers rarely would like selection methods to find large clusters of structurally similar compounds in the library (eg, 500
0 benzodiazepam derivatives would be undesirable). Singletons, ie compounds that do not have similar structures in the data set, are generally considered undesirable as they do not allow the opportunity to develop any structure-activity relationship information. Rather, a selection method that leads to a set of 10-15 small structures is considered preferable. Such a small set of similar compounds allows some analysis of the effects of small structural variations on the activity of the compounds (called structure-activity relationships, or SAR studies). Moreover, the small clusters help confirm the screening results. Given that 5 out of 10 compounds in the small cluster demonstrated biological activity, this cluster is compared to a chemically related structure and the activity is reproducible and "optimized". It is more likely.

Initial biological screens generally produce compounds called “hits” or simply “hits”. Hits are compounds that have been screened in a test and have demonstrated biological activity above a desired threshold. These hits may rarely include the final drug candidate for further analysis by animal toxicity studies and ultimately human clinical trials. In fact, these hits generally represent reads that have been optimized by making small changes in their chemical structure. This change is generally intended to improve or enhance the biological activity of the lead until additional screens identify candidate products. The compounds tracked are
It can be called the analog of the first hit. This process of hit optimization is commonly referred to as "lead tracking."

Lead tracking has generally been accomplished by chemical chemists who make small sets of analogs of several lead compounds. The biological potency of the analog is then tested, as well as the initial screen leading to the first hit. Structural changes that increase activity are chosen, those that decrease activity are usually discarded, and new modifications to the analog compounds are often made and tested. The medicinal chemist follows the lead until the compound (or small set of compounds) is confirmed to have adequate potency as a drug candidate.

In the last few years, pharmaceutical chemists have sought quantitative structure-activity relationships (Quanti
native Structure Activity Relations
ip, QSAR) and has often been supported by computer-based design techniques. These programs use the potency data for previously tested compounds to predict the potency of compounds still under test. The goal of the QSAR program is to give an accurate prediction of activity prior to testing compounds. The QSAR program is generally successful, not only predicting the activity of the final drug candidate, but also allowing a more efficient selection of analogs at each stage. Compounds predicted to be active by the QSAR method do not always have the expected activity, but in general these compounds have a greater chance of being active compared to others in general.

The development of pharmaceuticals is generally very competitive. Therefore, in most cases, once a drug candidate has been selected, extensive patent search is conducted to ensure that the candidate itself or its use is not limited by other patents. Animal toxicity searches generally follow patent searches. Where animal toxicity studies are acceptable, human clinical trials of drug candidates are conducted.

The process of screening for potential drug candidates, analog production and confirmation can be very time consuming and expensive. Searching for patents, especially in the area of drug compounds, is also time consuming and expensive. Animal toxicity studies, including potential drug candidates, cost thousands of dollars and are easy. Human clinical trials designed to establish the safety and efficacy of potential drug candidates in humans exceed $ 10 million. Therefore, a substantial input of time, effort and money may be due to potential drug candidates, for example within the scope of a third party patent, or other compounds and chemistries proven safe, for example by human clinical investigations. It is imperative to know as much information as possible about potential drug candidates as early in the process as possible so that they are not directed to relevant relevant drug candidates.

[0018]

[Relational database system]

Relational Database Systems (RDS) are widely used in all industries and academia to store and retrieve too much information of interest. RDS is
A table structure is used to store information about the different cases of each entity.
These tables define the columns that can be the attributes of each data item (column). The data in each column can be in several formats, including text, numbers, date and time, binary, and so on. The data in a column is indexed so that it can be retrieved more quickly.

In a relational model for database design, data that is repeated over sequences is usually split into new table / entity definitions. This process is called "normalization" and is generally done to protect data consistency and save space. However, the relationship between the data in the table is maintained.

Data in RDS is typically queried by users or by application programs by making specific queries in a query-oriented language. The Oracle ^™ system is a preferred example of RDS. In Oracle, as in many other RDS, queries are structured query languages (Structureare).
d Query Language, SQL). This language facilitates retrieval of information stored in various tables and allows combining relevant data in different tables. The composition of an SQL query that performs a combination of data is called a "join". The term "association" is a technical term in the text and is a noun, not a verb. A join concatenates the columns of one table and the columns of the other table based on some common or related column (attribute). The combination is "on the fly (on
can be pre-determined to provide a pseudo table, commonly referred to as a "view", or the join itself is added to each query as it occurs. Views have a new table look, but views are generally not stored as such.

[0021]

【the Internet】

The Internet is a useful technique for making information in different places available to different individuals. In fact, in vast Internet environments, individuals can access information tools remotely. The Internet is the preferred method for accessing information stored in relational databases such as those mentioned above.

The Internet, which began in the late 1960s, is a computer network made up of many small networks throughout the world. Internet host computers or computer networks allow the public access to databases containing information from experts in many fields. Hosts are sponsored by a wide range of entities including, for example, universities, government organizations, businesses and individuals.

Internet information is made available to the public via a server running on an Internet host. The server creates a document or file that is available to those who access the host site. Such files may be stored in a database, preferably located on the host, or a storage medium, such as an optical or magnetic storage device.

Networking protocols can be used to facilitate communication between the host and the requesting client. TCP / IP (Transmissio)
n Control Protocol / Internet Protocol)
Is one such networking protocol. A computer on TCP / IP uses a unique identification (ID) code that can uniquely identify each computer or host on the Internet. Such code may include an IP (Internet Protocol) number or address and corresponding network or computer name.

Created in 1991, World-Wide-Web (Web or www) allows users to intuitively navigate Internet resources without an IP address or other specialized knowledge. Provide access to information on the Internet while allowing. The Web comprises a large number of interconnected "pages" or documents that can be displayed on the monitor of a user's computer. We
Page b is served by a host running a special server. These We
The software running on the b server is relatively simple and available on a wide range of computer platforms, including PCs. Web browser software is equally available to display web pages as well as traditional non-web files on your system.

The Web is based on HTTP (Hypertext Transfer Protocol).
based on the idea of hypertext and transfer methods known as ol). HTTP
Is designed to run over a first TCP / IP and uses standard Internet setup, the server publishes or processes data and client representations. One format for information transfer is Hypertext Marku.
p is for creating a document using Language (HTML). HTML pages are made from standard text as well as code that indicates how to display the page. The browser reads these codes to display the page.

Each Web page can include pictures and sounds in addition to text. The pictures or sounds combined with some text connect to other computers on the same server or on the internet. This is known as a hypertext link. For example, links can be represented as underlined or highlighted words or phrases. Each link is a URL (Uniform Res)
It is directed to the web page using a special name called the Locator). The URL can go directly to the combined file, even if it is on another web server.

In addition to the Internet, which allows the general public to retrieve information, there are and commonly used alternative means of accessing such information. For example, direct modem connections between two computers, large academic societies and private Internet networks within an organization are equally available and useful means for accessing classification information stored in databases.

[0029]

[Outline of the Invention]

The present invention is directed to systems and methods for information organization in which features related to an entity can be inferred from features of smaller entities. According to one aspect of the invention, database similarity joins can be used to infer property or parameter related information contained in a first database from property or parameter related information contained in a second database. This aspect of the invention allows the invention to provide a search for information that is not organized in the manner requested or desired by a particular user. This allows searches based on common characteristics or similarities between organized entities in irrelevant databases. Information can be retrieved and organized in ways that make the information more useful to the user.

One method that allows such searching and organizing is called fuzzy similarity join. By this way, the relationships between the retrieved information do not have to be intuitively or systematically related to the way it is retrieved. Alternatively, the search for functional information can be based on similarities between entities in one or more databases.

These and other aspects of the invention, which may be carried out individually or collectively, are perhaps best explained in the application section. For example, 1
Or consider the application of drug discovery when you want to get some information about more compounds. With conventional drug database strategies, it is not easy or impossible to obtain information of interest to scientists in a database about a compound of interest. One aspect of the invention allows scientists to perform database joins to obtain information about compounds of interest from another database.

According to another aspect of the invention in this application, the scientist, based on the properties of the other parameters of the “similar” compound, may infer drug-like bond (or fuzzy-like bond) to infer information about the compound of interest. )It can be performed. According to this aspect of the invention, drug-like binding allows scientists to search one or more databases for information about "similar" compounds. Scientists can use this information to infer behavior or other properties or parameters for the compound of interest.

For example, in one embodiment, the drug space is defined such that there is a neighborhood effect on the property of interest (eg, toxicity), and then the properties of the compound of interest in one database are compared to those of similar compounds in another database. It can be inferred from the property database. Thus, this aspect of the invention in this application allows the joining of two tables by a similar comparison of the two structures. The exact fit of the two structures is
It is not required to do the combining work.

According to another aspect of the invention, a search tool can be used that combines actual compound data with a virtual data set to facilitate a neighborhood search around a preferred set of characterization metrics. Neighbor relations can make this the basis for similarity joins.

According to another aspect of the invention, datasets can be screened for exclusion of records that have special or identified characteristics or features. Further, the dataset can be combined with other data to allow further screening to eliminate unwanted classes of records.

According to yet another aspect of the invention, the search tool can be linked to an ordering system that allows a user to purchase a confirmed item.

These and other features, advantages and aspects of the present invention are described in further detail below.

[0038]

Detailed Description of the Preferred Embodiments

[0039]

[Introduction and overflow]

The present invention relates generally to information databases, and more particularly to database-like joins, and more particularly to systems and methods for organizing information so that features about an entity can be inferred from features of similar entities. The invention disclosed in this patent document provides information based on common features between entities that are needed or desired by a particular user, but which are organized into irrelevant databases (from data otherwise). It is applicable and useful for ordinary information retrieval that it is not organized in a way that can still be useful.

Here, a method that allows such retrieval and organization is called “fuzzy similarity combination”,
In this, the relationship between the retrieved information is not intuitively or systematically related to the way it is retrieved, but rather the relationship is a source of data that the user has (if possible) irrelevant or diffused. It is based on the user's request that it is necessary to painstakingly search for required data from In fact, the present invention differs from "fixed" catalogs (whether printed or electronic-based) that force the user to obtain information in a special and limited way according to the cataloger's goals. Thus, the present invention allows "fluid" retrieval of information based on the user's needs and purpose.

A fuzzy-like join may, for example, (1) identify at least one attribute that is common to the databases being compared, or (2) a set of operator-related decision sets (eg, equal to, greater than, Smaller than, include, exclude, etc.) does not include traditional joins that generally require organizing and retrieving information from a dataset based on fit. For example, the bonds in FIGS. 4A, 4B and 4C constitute a traditional bond and are not fuzzy like bonds. Fuzzy-like joins are based on non-traditional database operators. In this sentence, the nontraditional operator evaluates the similarity of two elements. "Similarity" is the composite index of some or all of the element attributes being compared, and the composite index is any value between 0 (not a perfect 0) and 1 or less (not including 1). Can be assigned to. For traditional concatenation, the composite exponent for the two elements must be exactly 1 (or will completely meet the conditions set by the standard association operators). For fuzzy similarity joins, the composite index should have a value greater than some value predetermined by the user. For example, in the preferred embodiment of the fuzzy-like bond disclosed herein (drug-like bond), the Tanimoto coefficient functions as a composite index of the drug-like bond, and in order to establish the drug-like bond, the user must: You will set a minimum value for the Tanimoto coefficient based on your particular needs.

Accordingly, while this disclosure focuses on certain specific forms of information, ie, specific examples directed to drug compound information, this disclosure, when viewed by ordinary skill in the art, does not disclose the present invention. Would provide opportunities to apply to other areas such as, but not limited to, biological compounds, metallurgical compounds, genetic information, health consciousness, population studies, political and public opinion trends, and the like. However, for purposes of presentation efficiency, and without limitation, the focus of the present disclosure is on certain fuzzy, drug-like bonds.

According to one aspect of the invention, a search tool combines actual compound data with a virtual data set to facilitate searching for neighborhoods around a preferred set of property metrics.

According to another aspect of the invention, the data set can be screened for compounds with particular properties or identified properties. In addition, the data set can be combined with other data to allow further filtering to exclude unwanted classes of compounds. For example, the dataset can be combined with information related to patent coverage, toxicology information, binding data, and the like. According to another aspect of the invention, the exploration tool can be linked to an ordering system that allows the user to purchase the identified compound.

The present invention and its various aspects, which may be implemented individually or collectively, are disclosed in the Internet term as preferred interface tools. The invention disclosed in these terms is provided solely for ease of explanation. After reading this disclosure, it will be apparent to one of ordinary skill in the art that the present invention may be implemented in any of several search environments.

[0046]

[Library information integration]

The most preferred embodiment of the information relating to library information integration, i.e. the integration of drug libraries, has been elucidated, which is made up in the terminology of the Internet environment as described above.

FIG. 1 is a block diagram generally illustrating one application of a library integrated information system according to one embodiment of the present invention. The application shown in FIG. 1 includes a library integration server 104 having access to libraries 106, 108 as well as another data source 110. In the preferred embodiment, and in the context of drug-like binding, library 106 comprises one or more drug compound libraries that are accessible and searchable by library integration server 104. The library 106 contains information about known and existing drug compounds and can also include information such as, for example, compound characterization metrics and other information. In one embodiment, where the library integrated information system may also be used to purchase compounds, the library 106 may include price, availability, and shipping information for the purchase of available compounds. .

Library 1 depending on the application of the drug library information system and the library 106 used in connection with the drug library information system.
06 may be located at or remote from server 104. The local library 106 can be used to provide direct access to the drug data contained therein and can be maintained in association with the drug navigation server 104. However, the compound library 106 may exist and be maintained by a third party, and may be accessible to the server 104 via a telecommunication network such as a dial-up link, a network, the Internet, or other communication medium. Under this scenario, the ability of server 104 to utilize additional datasets during operation can be extended to external datasets.

The one or more virtual compound libraries (only one shown) 108 preferably include identifications of virtual compound data sets that can provide known or available compound data sets. “Virtual” is a technical term that generally refers to drug compounds that may not be physically present. Virtual compounds can be defined by known synthetic processes that allow the synthesis of virtual compounds. Preferably, the virtual compound library comprises a plurality of virtual or hypothetical drug compounds that have a particular set of properties defined by one or more property metrics. Mapping or combination of virtual datasets with known datasets can be utilized to allow enhancements to compound search techniques, as described in detail below. Compound library 1
Similar to 06, the virtual compound library 108 can be maintained locally or at a remote location and can be accessed by the server 104 through a variety of different connection technologies.

A user who wishes to access the drug library information system can connect to the server 104 via the user's computer or workstation 102. User workstation 102 may connect to server 104 using a variety of post connection techniques including, for example, direct connection, network connection, or otherwise. In the preferred method, a user can access the server 104 from various remote locations via the Internet.

The third category of library shown in FIG. 1 is a library containing other data 110. Other data may be relevant to the drug compound, such as patent data, toxicity data, drug compound-biologic target binding data, or other data of interest to the user in evaluating the drug compound via the server 104. Can include data. Similar to the libraries 106, 108, the data contained in the library 110 may be found in a common database with other data, or may be found in a separate database that may be accessed locally or remotely by the server 104. it can.

Library 106, virtual compound library 108, and other data sources 110 are shown as “separated” databases in FIG. 1, but the data contained in these databases may be one or more physical databases. Alternatively, it will be apparent to one of ordinary skill in the art that it can be provided in a logical database. Further, although one server 104 is shown in FIG. 1, the system can be implemented using one or more servers 104.

The example application shown in FIG. 1 comprises at least two classes of libraries, a compound library 106 and a virtual compound library 108. As mentioned above, the library 106 contains information about known or existing chemicals or compounds listed in the library. Known drug compounds can be classified according to the values of a number of property metrics associated with each compound. According to one aspect of the invention, this catalog can be achieved by mapping compounds into drug space and combining known and virtual compounds in drug space.

[0054]

[Measurement standard and chemical space]

Before describing the mapping of compounds in drug space, a mathematical introduction to "space" is provided. As used herein, "space" is defined by a set of parameters and a distance function. The domain of parameters can be a set of real or complex numbers or any suitable sub-set thereof (eg, integer, positive integer, etc.). Parameters based on a discontinuous set of numbers are called "discrete parameters". A specific value for each parameter in space defines a point in the appropriate space. The distance function produces real, non-negative values for any two points in space.

The majority of individuals are physically related to the “available” life experience in space-2D (
We are accustomed to 2D) and 3D Euclidean spaces. The Euclidean space has a property that does not apply to all spaces, and in particular, the property of the Euclidean space does not apply to a space containing all drug compounds (drug space).

Drug space is understood to be similar to Euclidean space and is based on a space that can map some or all of the possible compounds. This space contains a very large number of compounds. The number of organic drug compounds with a molecular weight of 800 or less is estimated to be about 10 ²⁰⁴ .

The drug space is generally defined by the chemical structure of the compound and the parameters that can be calculated from the dissimilarity distance function based on these parameters. Parameters that can be used to define the drug space include, but are not limited to, for example:

CLogP Calculated (estimated) partition coefficient for the compound between octanol and water; molecular weight (MW); size related sterimol parameters; keel and related electronic properties of structure. Hall parameters; B-cut metrics as used in the Diversion Solutions ^™ software package (U. Texas); MDL, Tripos Inc. , And various fingerprint metrics used in software packages commercially available from Daylight Chemical Information Systems; Molecular Hologram metrics defined in HQSAR ^™ software commercially available from Tripos. The above are examples and do not mean exhaustive.

Unity fingerprint space and HQSAR hologram space (both T
ripos, Inc. St. Louis, MO. Is an example of the set of drug space concepts described above, eg Unity Fingerprint is an example of a synthetic drug space metric. However, the invention is not dependent on these spaces and is not limited to these spaces or metrics.

In the unity fingerprint space, the parameters are the various bits in the unity fingerprint bitmap. This bitmap is
For each chemical structure, it is created by breaking the structure into all of its substructure fragments. The unique representation of each fragment is then recreated at some position in the bitmap. Fingerprint bitmaps are typically 200 to 1000 bits long. When a fragment is found in the structure, the bit that maps it is 1
Is set to. Therefore, the fingerprint is a series of bits that indicates the presence of all of the fragments in the structure.

The same bit should always be set to the same bit in the fingerprint, regardless of how it is encountered. So two structures that have many of the same fragments will have many of the same bits set in their corresponding fingerprints.

For example, when the 1000-bit fingerprint is used, the drug space thus defined is a 1000-dimensional discrete space, and each parameter is 0 or 1.
Can be either. This space is a 1000 dimensional "hypercube" and each possible drug maps onto one of the vertices of this hypercube.

The distance function of this space is based on one of the similar indices ranging from 0 to 100%.
The distance is 1 minus the value of the similarity index (d = 1-sim). Common similarity indices include the Tanimoto index, the cosine index, and some others.

[0064]   The Tanimoto index is defined as follows.

Sim = C / (A + B−C) where C: the number of bit sets in both fingerprints (= 1) A: the number of bit sets in the first fingerprint B: the number of bit sets in the second fingerprint Is. The Tanimoto coefficient is used almost universally, and in applications the coefficient is commonly associated with the chemist's intuition regarding compound similarity.

For simplicity of description, the drug space is referred to as a 2D Euclidean space in this document. This is done for convenience only. The drug space defined by the unity fingerprint and the Tanimoto index is not Euclidean space, but is slightly discrete according to the triangular inequality.

All compounds within a distance (ie, having a small number of similarity indices) in the drug space define a “vicinity” of the compound. These compounds are said to be within the similar radius of a compound or within the sphere of the compound's neighborhood. In non-Euclidean space, a sphere is defined as every point within a distance of another point. Non-Euclidean spheres do not necessarily have the properties of spheres in 3D Euclidean space, have no defined surface or surface area, and have no volume. The neighboring sphere in the figure is a 2D projection of the sphere (circle) here for convenience of representation.
Will be represented by

[0068]

[Two-dimensional representation of multi-dimensional drug space] The concept of drug space was introduced, but here an example of mapping compounds into multi-dimensional drug space is given.

FIG. 2 is a diagram showing a mapping of an example subset of known drug compounds in a hypothetical 2D drug space. The actual drug space is multi-dimensional and contains multiple metrics rather than two metrics (metric 1 and metric 2). In FIG. 2, compounds are mapped in this space based on the compound's property values for two property metrics.

There are drug database systems that map compounds into drug space according to property metric values and further define neighborhoods for compounds according to property metric values. Examples of such databases include Tripos's Unity database system MDL's Isis ^™ database system, and others. One of ordinary skill in the art will appreciate that these commercially available databases can be used as one or more libraries 106, or custom libraries can be created for a given application. It will be apparent after reading.

Further, while FIG. 2 shows a combination of two libraries, one or any number of libraries can be combined in a given drug space for navigation. Library 1 because the population of a given compound is dynamic and constantly changing
It is conceivable that 06 can be updated and changed to reflect changes in the chemical community. Further, as in this exemplary drawing, many different drug spaces can be defined and all possible drugs can be mapped into each of these spaces.

In FIG. 2, the compound is represented by points 122 and circles 124. Point 122 and circle 12
The use of 4 is a database of known compounds mapped to the drug space, one database of compounds represented by points 122 and another database of compounds represented by circles 124, two different databases. Show that you can get from. Compounds from additional databases can also be mapped into this drug space, as will be apparent to those of ordinary skill in the art having read this specification. Mapping a compound from one or more databases into a multidimensional space and representing up to N characterization metrics for this compound, as will be apparent to those of ordinary skill after reading this disclosure. You can

Various drug neighborhoods 126 that create a 2D drug space are also shown in FIG. "Cell base"
The neighborhood 126 is shown in FIG. 2 bounded by a dashed line. The cell base neighborhood is defined by partitioning the metric into several "boxes". This method can lead to cell base neighborhoods containing boxes that do not contain any compounds. Compounds 122, 124 that fall within a given cell base neighborhood can be considered similar compounds.

Alternative preferred neighborhoods are based on points within a region or distance of a given compound or virtual compound, as will be described with reference to FIG. 3 and large circle 132. Compounds within circle 132 are said to be in the vicinity of virtual compound 130A, which contains actual compounds 122A and 124A. This "distance-based" neighborhood can be defined by actual or virtual compounds, each compound defining its own neighborhood.

FIG. 3 shows an example of a virtual compound library in drug space combined with a library of known drugs as shown in FIG. In one embodiment of the present invention,
The virtual library 108 contains a large amount of virtual compounds compared to the number of compounds in the library 106, and preferably the virtual compounds are relatively evenly distributed over the drug space. As a result of this execution, it is possible to find a virtual compound that fits well with the given set of characteristic measurement reference values, regardless of the presence of previously synthesized compounds having characteristic measurement reference values.

As shown in FIG. 3, there are a relatively large number of virtual compounds 130 compared to known compounds 122, 124. Due to the large number of virtual compounds, the virtual compounds can occupy multiple positions within the defined drug space. So, from a probabilistic point of view,
It is more likely that the user will find a virtual compound that corresponds or fits well into a given set of values for the desired property metric specified by the user. The combination of the virtual data set and the known data set allows the virtual compounds to be grouped into neighborhoods by known compounds.

[0077]

[Nearby effect]

A drug space can be considered useful if it exhibits a "neighborhood effect" for certain relevant properties. The neighborhood effect is when a compound that is similar to a particular compound with a particular desired value is more likely to have a property value that is more similar to the property value of a particular compound than the population of all compounds. Occurs. Using Unity Fingerprint and Tanimoto's Similarity Index as an example of a neighborhood effect, a compound that is 85% similar to an active compound may be 30 times more biologically active than a randomly selected compound It was found.

The drug spaces found to create a neighborhood effect on biological activity and toxicity are the b-cut space of the divers solution (U. Texas), the fingerprint space (Tripos, MDL, Daylight) and Cerius2. (
SAR space of MSI). However, it should be noted that certain properties and metrics do not produce near-field effects. For example, for drug spaces based on molecular weight and cLogP, it is not expected to find potentially potentiated active compounds in the vicinity of the active compound that are also biologically active. Therefore, the molecular weight and c
The space defined by LogP does not create a neighborhood effect on biological activity.
Those of ordinary skill in the art believe in the ability to establish neighborhood effect criteria that suit the special needs of the skilled person.

[0079]

[Fuzzy-like bond, drug-like bond]

Drug-like bonds can be used to help identify compounds that may be of interest to the user. But before revealing drug-like bonds,
It is useful to first explain the concept of join operations in relational database systems.

In relational database systems, tables are used to store records of relational information. Each table represents an occurrence of the entity type, so the attributes of the table define the entity. For example, a table may be used to store information about various departments of a company. The attributes of the table may include items such as department ID number, department name, budget code, manager name, building location, and so on. An example of such a table is shown in Figure 4A.

Another table may include information about company employees including employee ID number, social insurance number, name, department ID number, office location, job title, and so forth. An example of such a table is shown in Figure 4B.

The information in these tables can be combined to create information about the subject using database “joins”. A join combines the columns of one table with those of another table based on the relationship between the values in one column of the record. For example, the employee table can be combined with the department table to provide a new pseudo-table (called a view) that gives the department name for each employee. An example of a Structured Query Language (SQL) collation that would achieve the creation of a join is from the department table, employee table, (department table department ID = employee table department ID) Name You can choose to select name, office, employee ID, department name. A view table created using such SQL matching is shown in Figure 4C.

Relational databases can be used to store information about various compound libraries in a similar manner. In the context of drug compounds and related information, different libraries may have different attributes, including chemical structure, vendor, price, location, toxicity data, bioscreening data, and so on. Information contained in various libraries can be combined to increase the value of the information. For example, a table contains data available from the merchant A, the second table may contain data of Rat LD ₅₀ of the set of compounds. These tables can be combined to provide information on compounds that are available and have good toxicity results. In one example, this attachment can be accomplished using a chemical structure.

FIG. 5A is a diagram showing an example of a vendor table for compounds. 5B is the same as FIG.
FIG. 6 is a diagram showing an example of a toxicity table for a subset of the compounds listed in Table A.
FIG. 5C is a resulting binding table showing availability of compounds in the table of FIG. 5A and Rat LD ₅₀ data for the set of compounds in the table of FIG. 5B.

The bonds shown in FIGS. 5A-5C are useful in situations where the structures in the available catalog have been tested for attributes of interest. The examples of toxicity data given above are useful if they are in the toxicological endpoint and the "toxicity" table in which this data exists. Unfortunately, to date, very few compounds are available from vendors with known toxicology data, or from chemists who produce (link) a lot of data. Thus, it is unlikely that the user will obtain toxicity information about the compound of interest unless the user searches for a particular compound that has been tested for toxicity using a traditional drug search catalog. This is the first reason why drug-like bonds are an essential feature of the present invention.

In this context, this disclosure of drug-like bonds is an example of our disclosure of “fuzzy” like bonds. In a broader context than the example situation, the fuzzy like join is
By following the same logic, that is, organizing or combining information that is not normally combined or organized together based on similar properties between compounds within library information integration.

[0087]

[Genomics]

A similarity index is defined for the two sequences based on the number of stored bases or on a more sophisticated analysis of the sequences as exemplified by the FastA or Blast search system. This similarity measure based ligation is used to connect sequences of unknown genes or unknown functions using a well-characterized database of genes. This helps place the new gene within a family or superfamily, or helps identify its function.

[0088]

[Group clinical trial]

Candidates for clinical trials include standard demographic data, physical characteristics, personal profile tools,
It can be described by various measurement standards including driving records. The results of gene profiling can also be used as descriptive words in "people space".
The distance measure can be determined in several ways, and it can be characterized as the magnitude of the distance between any two persons mapped to this people space. An example of the distance function is as follows. 1) Normalize each people space metric to the range 0-1 based on the minimum and maximum values in the entire set. 2) Calculate distance by

D (A, B) = Σmin (A ₁ B ₁ ) / max (A ₁ B ₁ ) This is a general form of Tanimoto coefficients for spaces not limited to parameter values 0 and 1. This similarity measure can be used to link a database of potential subjects to a database of previous test results in which a person has experienced undesirable side effects. The sensitivity of the new subject can then be estimated from the previous results based on similarities to the previous test subject.

Fuzzy-like joins, including examples of drug-like joins disclosed herein, include, but are not limited to, relational database systems, unrelated database systems, file-based information systems, spreadsheet-based systems, or indexes. It can be implemented in a variety of ways, including cards, non-computer based systems including collection of reports, etc. The preferred embodiment will use the Oracle database system. In Oracle, a fuzzy database join can be implemented as an Oracle cartridge that defines a chemical structure as a new entity and a similar comparison as a new operator action of a chemical structure entity. The combination can be performed in Oracle as an external procedure cell. In this embodiment, the SQL language is enhanced to specify a match to allow fuzzy like join specifications and other matching criteria.

Another implementation may include pre-calculation of fingerprint bitmaps or other structure-based metrics and storage of the pre-calculated metric within Oracle or in a file of the Oracle external system. Yet another implementation would include similar precomputation of some or all pairs of structures and storage of similar values in oracle or in an external file or system.

The present invention allows "ideal" combining of records (ie records are not combined in the desired way) based on fuzzy similarity, whether or not using a relational database system such as Oracle for implementation. Relevant and independent of the integration of fuzzy-like join matching specifications with other matching criteria.

Databases that would reinstate the drug-like bond and enhance the information values in the first order table for exactly the same chemical structure are often unavailable, but similar structures are available Note that there are many. The drug space is a characteristic of the problem (eg,
The properties of the compounds in the first order table can be inferred from the property data for similar compounds in the second order table, provided that there is a near effect on toxicity.
This is because, for example, two tables are not exactly structurally compatible, but can be joined by a similarity comparison of the two structures, which is the basis for drug-like binding.

In a drug-like bond according to one embodiment, if the similarity index of two compounds is greater than a certain threshold (eg, 80%), one row of the two tables is obtained.
Combined in rows. Drug-like bonds for which there are neighborhood effects of property examples and are therefore useful include, but are not limited to:

[0095]       Biological effects-a table (all active) made from hits in the primary screen       To find the SRA neighborhood to be tested in the secondary screen,       Can be combined with those in the available medication table.

[0096]       Toxicity data-table of available compounds should be combined with table of toxicology data       You can The table obtained is for each compound in the following table       Average toxicity data for the neighborhood can be included. This value is converted to the primary table       Estimate of toxicity for compounds, removal of potential toxic compounds, or testing       Can be used to prioritize trial schedules.

[0097]       Patent Scope-Some patents have a specific structure and a comprehensive description of the drugs in scope       May be included. Specific structures available in database of patented compounds       Often possible. Determine if compound is within comprehensive description       It is more difficult to do, but can be done one by one. Obtained by the patent table       The drug-like bond of the obtained drug table is directly related to the scope of patent and has several relations.       It can be used to create a good structure. This is the "patent warning flag"       Can be represented, for example, the 20 neighbors of the compound can be represented by one or more       If within the scope of the patent, this is whether this compound is also within scope.       The user can be signaled to look up.

[0098]       Extensibility-a set of obtainable compounds (can be made but never made)       When a compound set) is combined with a large virtual database,       Estimate information about peripheral expansion capabilities. This is the primary screening       Select library elements or select from hit table of structures to be followed.       Useful for certain purposes. In this case, there is at least some proximity to the "synthesizable" drug space.       Compounds with aside are more valuable than those without. The number of virtual neighbors is       It is a measure of expandability.

[0099]       SAR primary selection-when a set of attainable compounds is combined with itself       (Regressive drug-like bond) is the number of neighbors each compound in the set has       Can be an attribute. As a result, the selection of compounds is small SAR.       A primary screening SET containing clusters of types can be formed.

The basis for drug-like bonds has been elucidated, but now the functional flow diagram of FIG. 6 showing the hypothesis for identifying potential compounds for users using drug-like bonds. Refer to. In step 252 of FIG. 6, a user seeking a compound that meets their needs can identify the target compound for which they are seeking.
The user will generally identify a particular compound of interest by its chemical structure (although other methods, such as chemical names, are available).

Since the user desires to look for a compound similar to the identified target compound, the user, in step 256, selects an acceptable neighborhood of values for the characterization criteria of the target compound identified in step 252. Define the range. The neighborhood can be considered as the range of the chemical space surrounding the target compound and the range in which the search is acceptable. Alternatively, as described above with respect to FIGS. 2 and 3, this step is not required so that the user may simply choose to search within a predetermined neighborhood (eg, a cell-based neighborhood where the compound is). .

At step 260, the user provides the server 104 with the target compound and the neighborhood. These data are submitted to the server 104 via the Internet. To facilitate the user in providing this information to the server 104, the format in this embodiment and other suitable interfaces may be provided to the user.

At step 264, drug-like binding is performed. Various techniques can be used to form the basis for drug-like bonds. For example, the Tanimoto coefficient can be used to determine the similarity between compounds and thereby whether a compound is within a selected neighborhood. Molecular holograms can be used to compare two or more molecular structures to determine if a compound is within a defined neighborhood. Molecular holograms and their use in QSARs can be found in Hurst et al., US Pat. No. 5,751,605, incorporated herein by reference.

At step 272, the combined records are averaged, counted, or processed by other statistical techniques to estimate the properties of the compounds in the primary data set.
For example, if toxicological data is included in the secondary data set, the average toxicity value of related compounds can be used as an estimate of toxicity for the primary compound. Alternatively, detailed information about the associated structure and its property values may be returned in its entirety.

At step 276, the server 104 provides the search results to the user at the user workstation 102. As described above, the user can evaluate the compounds returned as search results and decide whether to purchase the compounds for their research purposes. In one embodiment, the server 104 can coordinate or even handle the sale of the compound to users. For example, in the internet embodiment, the server 104 may also have the ability to complete the sale of the compound as a regular "e-commerce" site. Further, the user can go to another server or another site to purchase and complete the compound.

To illustrate various features and aspects of the present invention, some user scenarios are described. These scenarios describe use for one or more embodiments of the invention that may be utilized in a research setting. After reading this scenario, one of ordinary skill in the art will become apparent how to practice the invention for these and many other scenarios.

FIG. 7 is an operational flow chart showing an example of a scenario that allows a user to use a drug library integration tool to perform a primary selection of compounds from a large library or database of available compounds. Can be used. First, in this setting the user may wish to remove any unwanted compounds from the set of compounds under consideration. This is indicated by step 422. At this stage, compounds in the set that meet user-defined criteria as unwanted compounds are eliminated. For example, the user may use the molecular size, cLo
Exclusion of compounds can be determined based on gP range, reactivity or toxic functional groups, and the like. Such exclusion is based on the specific needs of the user.

At step 424, the user selects a neighborhood radius to define the similarity range he wishes to search. In one embodiment, the user chooses a radius for the SAR cluster. This radius is typically chosen to be about 0.8 or 80%, although other radii can be chosen.

In one scenario, the user may have small S in various areas within the drug space.
You can search for AR clusters. For example, the user can search for 10 compound clusters. Thus, in step 426, the user selects the desired neighborhood occupancy and also indicates the desired total number of compounds desired in the search process.

At step 428, the library or database is attached to itself using, for example, a chemical analogy. This bond creates a number of near neighbors for each compound. Once again, this bond creates a number of compounds within the defined neighborhood of each compound. The drug-like bond allows the user to seek out compounds that are within the selected similarity radius.

Compounds within various neighborhoods are counted, and compounds with at least X neighborhoods are selected with or near X and excluded from the data set. This selection is made from the set that was not previously excluded or selected. The remaining neighborhoods are excluded from the dataset. This is indicated by steps 430, 431 and 432.

This process of neighborhood exclusion is continued until the desired number of compounds has been selected for evaluation. This is indicated by step 434. Once the desired number of compounds has been selected or there are no more compounds to select, the corresponding person can order a sample of the selected compounds, eg for biological screening.

Another common run in the library is read tracking. This method
It is commonly used to determine if a lead is sufficiently important to give further consideration when the user follows a lead that falls outside of the broader search. FIG. 7 is a diagram illustrating an example method of using the present invention to perform lead tracking, according to one embodiment of the present invention.

Referring now to FIG. 8, a table of hits from the high power screen is shown, described, or loaded into the table. This is indicated by step 462.

At step 464, the user selects the desired similarity radius. As with the example scenario above, one similar radius used in this process is typically 0.8 or 80%, although other radii can be chosen.

At step 466, the hit table is combined with the table of toxicology data using drug-like joins. Because drug-like bonds are used instead of standard bonds, this bonding task involves compounds in the poison table that are within the similar radius of the subject compound. Once again, if the identified compound is found in the toxicology table, no subsequent conjugation work is necessary. Similar binding is more extensive, trapping compounds within a defined similar radius. In this binding, the toxicant data for the various columns that were bound are averaged to establish a prediction of toxicity for the primary table structure.

At step 468, the user selects a toxicity block value. For example, the user may undesirably seek no more than a defined minimum of a single dose of toxic drug. In step 470, hits with predictive toxicity values above the block value can be eliminated. That is, in one embodiment, a compound is excluded if the drug-like bond between the compound of interest and the toxicant table results in a small, one-time predicted dose of the toxic drug.

In this scenario, the user also considers the information subject to the patent coverage of the potential compound. In step 472, the resulting table is combined with the table of patented compounds to determine the likelihood of patents in the field. That is, the compound remaining after step 470 is combined with the proprietary compound table using a drug-like bond. In this combination, the number of combined records is counted. At step 474, the user enters the number of patent records that define the acceptable limit. The resulting table is examined to determine the number of compounds within the similar radius of the subject compound that are within the scope of the patent or affected. In step 476, hits above the cutoff value selected by the user may be eliminated.

At step 478, the user selects a tracking set. For example, the user can select about 10 compounds from the compounds remaining after the above-described similar binding to perform a broader test. At step 480, the selected compound is bound to a library of obtainable drugs. Again, the compounds are attached using drug-like bonds. In one embodiment, this coupling can be accomplished using an inner coupling without averaging or counting. At step 482, the combined structures are ordered, eg, for secondary screening.

As these scenarios show, one advantage of using drug-like bonds is that data about drugs or compounds within similar radii can be combined with target compounds to which similar bonds are made to behave, attribute, or otherwise Can be used to predict or infer the parameters of For example, the user may wish to have information regarding toxicity and patent coverage for a given compound or hypothetical compound. However, this data may not exist for that particular compound or hypothetical compound. Therefore, standard relational database join operations cannot provide the user with information about these parameters. However, the user can use drug-like bonds to perform binding tasks within a similar radius to obtain and explore information that may be relevant to compounds within a defined similar radius. At this time, the user
This information can be used to estimate or predict whether these parameters are present in a selected compound or hypothetical compound.

The various embodiments, aspects, and features of the invention described above can be implemented using hardware, software, or combinations thereof, and further using a computing system having one or more processors. Can be executed. In fact, in one embodiment, these elements are implemented using a processor-based system capable of performing the functions described above.

Although various embodiments of the present invention have been described above, it should be understood that they have been given by way of example and in a non-limiting manner. Therefore, the spirit and scope of the present invention should not be limited by any of the above-described exemplary embodiments, but should be defined only by the claims and their equivalents.

[Brief description of drawings]

FIG. 1 is a block diagram generally illustrating an application of a library information system according to an embodiment of the present invention.

FIG. 2 is a diagram showing “cell-based” neighborhood mapping of an example subset of known compounds in a two-dimensional representation of a multi-dimensional drug space.

FIG. 3 is a diagram showing a “distance-based neighborhood” mapping of an example data set of known compounds to an example data set of virtual compounds according to one embodiment of the invention.

FIG. 4A is a diagram showing an example of a table that may be used to store information in various departments within a company.

FIG. 4B is a diagram showing an example of a table that may include information about company employees including employee ID numbers, social security numbers, names, department ID numbers, office locations, job titles, and the like.

FIG. 4C is a diagram showing an example of a table that can be obtained from an example SQL that matches database joins.

FIG. 5A   FIG. 6 is a diagram showing an example of a vendor table of compounds.

5B is a diagram showing an example toxicity table for a subset of the compounds listed in the table of FIG. 5A.

5C is a synthetic linkage table showing the availability of compounds in the table of FIG. 5A along with Rat LD50 data for the set of compounds in the table shown in FIG. 5B.

FIG. 6 is an operational flowchart illustrating a powerful compound identification process for a user according to an embodiment of the present invention.

FIG. 7 is an operational flow diagram illustrating an example scenario for primary selection according to one embodiment of the invention.

FIG. 8 is an operational flow diagram illustrating an example scenario for a tracking lead according to one embodiment of the invention.

[Procedure amendment]

[Submission date] March 20, 2002 (2002.3.20)

[Procedure Amendment 1]

[Document name to be amended] Statement

[Name of item to be amended] Claims

[Correction method] Change

[Contents of correction]

[Claims]

─────────────────────────────────────────────────── ─── Continued front page    (81) Designated countries EP (AT, BE, CH, CY, DE, DK, ES, FI, FR, GB, GR, IE, I T, LU, MC, NL, PT, SE), OA (BF, BJ , CF, CG, CI, CM, GA, GN, GW, ML, MR, NE, SN, TD, TG), AP (GH, GM, K E, LS, MW, SD, SL, SZ, TZ, UG, ZW ), EA (AM, AZ, BY, KG, KZ, MD, RU, TJ, TM), AE, AG, AL, AM, AT, AU, AZ, BA, BB, BG, BR, BY, CA, CH, C N, CR, CU, CZ, DE, DK, DM, DZ, EE , ES, FI, GB, GD, GE, GH, GM, HR, HU, ID, IL, IN, IS, JP, KE, KG, K P, KR, KZ, LC, LK, LR, LS, LT, LU , LV, MA, MD, MG, MK, MN, MW, MX, NO, NZ, PL, PT, RO, RU, SD, SE, S G, SI, SK, SL, TJ, TM, TR, TT, TZ , UA, UG, US, UZ, VN, YU, ZA, ZW

Claims (28)

[Claims] 1. A computer-based method for identifying one or more compounds having one or more similar properties, the method identifying a property of a target compound and having properties similar to those of the target compound. A method comprising using a computer to perform similar binding to identify one or more database compounds. 2. A computer-based method for identifying at least one item having at least one characteristic similar to the target item from a database containing the items, the target item being identified. And a method comprising using a computer to perform a fuzzy similarity join in a database to identify at least one item from a database that has properties similar to those of the target item. 3. The method according to claim 10, wherein the fuzzy-like bond is a drug-like bond and the item identified as the target item is a drug compound. 4. The method according to claim 10, wherein a computer user is informed of the identification of one or more identified items having characteristics similar to those of the target item. 5. The method according to claim 11, wherein the characteristic comprises a chemical structure of the target item. 6. The method according to claim 10, wherein a plurality of characteristics of the target item are identified. 7. The method according to claim 11, wherein a plurality of characteristics of the target item are identified. 8. One or more of the plurality of properties is a chemical structure, synthetic pathway, binding data, biological activity, structure-activity relationship information, molecular weight, partition coefficient, charge, size. 16. The method according to claim 15 selected from the group consisting of: efficiency, toxicity, manufacturer, price, and availability. 9. The method according to claim 12, wherein the user receives the information via a telecommunications network. 10. The method according to claim 17, wherein the telecommunication network is the Internet. 11. The method according to claim 11, further comprising excluding test items from the database by selecting user-defined criteria for undesired items. 12. The method according to claim 10, wherein the target item is a biological compound. 13. The method according to claim 20, wherein the biological compound is a protein. 14. The method according to claim 21, wherein the target item is a gene. 15. A computer-based method for identifying from a database containing drug compounds at least one drug compound having at least one property similar to the target drug compound, the method comprising: A method comprising using a computer to perform a drug-like bond in a database to identify a property and further identify at least one database drug compound having properties similar to those of the target drug compound. 16. The method according to claim 23, wherein the user is informed of the identification of at least one database drug compound having properties similar to those of the target drug compound. 17. The method according to claim 23, wherein the properties of the target drug compound include the chemical structure of the target drug compound. 18. The method according to claim 23, wherein a plurality of properties of the target drug compound are identified. 19. The method according to claim 23, wherein the characteristic identified is a neighborhood effect. 20. The method according to claim 27, wherein the neighborhood effect comprises a range of values of the characterization metric for the target drug compound. 21. The method according to claim 23, wherein the similarity between the properties of the target drug compound and the database drug compound is determined using at least one parameter selected from the group consisting of Tanimoto coefficients and molecular holograms. 22. The user receives information via a telecommunications network.
By the method. 23. The method according to claim 30, wherein the telecommunication network is the Internet. 24. The method according to claim 23, further comprising excluding at least one database drug compound from drug-like bindings by selecting user-defined exclusion criteria for unwanted compound features. 25. The chemical properties are chemical structure, synthetic pathway, binding data, biological activity, structure-activity correlation information, molecular weight, partition coefficient, charge, size, efficiency, toxicity, manufacturer, price, 24. The method according to claim 23 selected from the group consisting of: and availability. 26. A computer-based method for identifying from a database containing biological compounds at least one biological compound having at least one property similar to a target biological compound. Perform fuzzy-like joins in a database to identify a property of the target biological compound and further identify at least one database biological compound that has properties similar to those of the target biological compound A method that includes using a computer for. 27. The method according to claim 34, wherein the biological compound is a protein. 28. The method according to claim 34, wherein the biological compound is a gene.